I don't want to discourage new projects but why in the hell anyone would need Ai assistant on router? What's the use case?
1 Like
If you buy ASUS router you will get AI security checkup, not much AI, but indicates parameters non-proficient user may want to flip off or set better passwords etc…. Minimal neuron to indicate parameter other users were getting wrong would be nice (but precise audit script could work that out too)
I bet they spent megawatts of energy to train that model 
1 Like
I support this. As an SRE engineer who has been using OpenWrt for nearly a decade—both at my home and my parents’—I still feel I haven’t fully grasped its deeper aspects. When I encounter issues, I usually aim for solutions that minimize future maintenance and avoid disrupting OpenWrt’s upgrade path. This often requires me to think carefully about the development philosophy and roadmap of the OpenWrt team. So, beyond searching online or digging through official documentation (which takes considerable time), I hope there’s a more immediate and recommended approach that can directly point to official docs and OpenWrt’s future development vision
From an operational perspective, OpenWrt is already sufficiently user-friendly. The real challenge lies in the transition process of continuous upgrades—adopting new solutions while abandoning old ones—which requires additional effort. That said, I believe the OpenWrt team has worked hard to make this as smooth as possibl
So my expectation for this type of OpenWrt AI tool is that it can closely follow the design philosophy of the OpenWrt team, provide solutions that align more closely with the official approach, and directly offer official documentation to save time on active searching. I think at this stage, the best it can do is to provide practical and actionable suggestions. As for operations that deviate from official considerations, such as unsupported virtual machines and more niche container runtimes, they should be restricted and subjected to more serious scrutiny.
Another point is that the concerns about AI recommendations being unsafe, as perceived by others, actually have a relatively simple solution:
Since the data is stored locally, there is no need to worry about privacy issues. Before providing suggestions, the AI should first consider the context (such as OpenWrt's local configuration files and OpenWrt version, etc.) and then offer feasible recommendations
Well, thats not AI per se, you will not get local clippy talking into your head
I don't think anyone mentioned privacy issues. It seems that you did not read the thread carefully. The concerns voiced so far were about AI suggesting changes that may cause security, stability or preformance issues, and the vast majority of users not being able to properly validate such suggestions.
I do like your idea of AI assisting users with documentation, though. This could be actually a legitimate and useful feature (instead of AI directly suggesting changes).
This is a neat project, I can see it being useful from a research and learning perspective immediately, and with some guardrails as a more active assistant as it looks like you originally envisioned.
It's interesting - the posters responding in this thread bring up valid concerns but also remind me of early opponents of other now foundational tech. “AI” and LLMs are clearly here to stay, it's strange that even in this sort of community you see people bashing another person exploring the art of what's possible; what's worse, many of them with vague or only tangentially relevant arguments one would moreso expect to see from talking heads on TV rather than in a technical forum. It's like someone having qualms with HTTP as a protocol yet they criticize someone building a website with it. You're just shooting the messenger.
Done rambling - keep up the good work!
This claim in itself is vague. Concerns about users receiving bad advice for configuration of an important system are very specific and there is nothing vague about them.
You are making assumptions about people's intentions. Bad idea. Who here said that they oppose AI? Again, the voiced opposition was against bad config recommendations provided by AI, which currently is pretty much a statistical inevitability, as was shown multiple times in this very forum. If and when AI-generated advice becomes reliable, this stance will change - but things are not there yet.
1 Like
There could be a privacy concerns if the user transmitted configuration parameters or identifying specifications (e.g. asking AI to create a Wireguard tunnel - providing the private key, peer information, etc.).
This is more responsibility of the user to maintain thier security, though.
1 Like